sem: improve and rework semobjconstr

[zerbina]

Summary

Rework the semantic analysis of nkObjConstr nodes. The logic is now
simpler, compile-time evaluation happens from left to right, non-follow-
up errors are not dropped, and less follow-up errors are reported.

Details

The idea is to split the processing into two phases:

1. AST validation and semantic analysis (i.e. figuring out the meaning)
2. filling in the gaps (e.g., fitting types) and making sure language
   rules are not violated

Phase 1

During the first phase, for each slot, it is verified that the AST has
the expected shape and that a field with the given name exists in the
used object type; the initializer expression is also semantically
analyzed.

If the AST for a field assignment is not as expected, it is not analyzed
further. Compared to the previous logic, the "invalid field assignment"
error is not specialized further -- this is left to error message
rendering (currently astDiagToLegacyReport), thus making
adSemFieldAssignmentInvalidNeedSpace obsolete.

Errors in phase 1 do not short-circuit the second phase.

Phase 2

The second phase is split up into two parts:

1. post-processing of the initializer expression and basic validity
   checks that only need the object construction expression
2. validity checks that also need to consider the object type

The post-processing includes (if possible) folding the initializer for
discriminators and fitting the expression's type to the formal type; the
basic validity checks are the field visibility and duplicated field
checks. Errors nodes produces here are placed directly into the
temporary AST.

This allows for dropping the semExprFlagDispatched procedure and the
efPreferStatic, efPreferNilResult, and efPreferStatic expression
flags.

Running the checks that require traversal/inspection of the object type
re-uses the pre-existing semConstructTypeAux and semConstructFields
(both which have their sem prefix replaced by check). Both only
inspect the construction expression now (no sem-checking or modification
is performed) and rely on the field assignments that reach them to either
be an nkError node or valid, which significantly simplifies error
handling.

Since it is obsolete now, the rsemFieldOkButAssignedValueInvalid is
removed. A specialized wrapper for incorrect object constructions is not
needed (the generic wrapper suffices), so
adSemObjectConstructorInvalid is also removed.

[saem]

Just an initial observation, this separates expansion/synthesis (phase 1) and reduction/inheritance (phase 2). Definitely reads a lot cleaner and the amount of lines removed is great.

I'm doing a trace of the logic just to make sure I follow along, but a big fan thus far. 🥳

[saem]

Much improved. I followed along the logic and it seems to cover all the bases, nicely done.

[saem]

/merge

[github-actions]

Merge requested by: @saem

Contents after the first section break of the PR description has been removed and preserved below:

---

To-do

• add tests for the changes
• write a proper commit message

Notes for Reviewers

• a small side-project; the complexity in semobjconstr has been bothering me for a while
• I wanted to use ElaborateAst here, but it doesn't work well for semObjConstr